Our research
The ultimate goal of our research is to understand how molecules interact in biological systems and use this information to accelerate the discovery of novel therapeutics. We develop tools to improve ligand discovery, leveraging computational chemistry and machine learning to interrogate large chemical libraries against the structures of clinically relevant targets. We also apply these tools to discover new starting points for drug design against challenging and unconventional binding sites. We take a collaborative and multidisciplinary approach by working side-by-side with groups with expertise in medicinal chemistry, biology, and pharmacology. Feel free to contact us for more information or to discuss collaboration opportunities.
Our codes are available on GitHub.
Accelerating the exploration of the drug-like chemical space
While early-stage drug discovery screening campaigns have traditionally focused on chemical libraries from thousands to few millions of molecules, it has been estimated that there are more than 10^60 possible drug-like molecules. To bridge the gap, recent efforts to develop virtual make-on-demand libraries led to the availability of enormous collections (>1 billion) of tangible small molecules, presenting unprecedented opportunities but also challenges for computer-aided drug discovery. In this context, we aim at developing new methods to alleviate the challenges of virtual screening at a large scale, by leveraging machine learning to build fast and reliable surrogates of physics-based methods. These new tools can improve accessibility to large chemical libraries and thus accelerate the exploration of their therapeutic potential. Read more about the topic in our review.
Improving structure-based virtual screening by combining data-driven and physics-based approaches
As the size of chemical libraries continues to expand, a corresponding increase in artifacts arising from virtual screening is observed. It becomes also increasingly challenging to identify ligands for novel targets lacking confirmed binders. It is also difficult to target proteins that undergo large conformational rearrangements upon ligand binding. Our research aims at marrying data-driven approaches with physics-based tools in order to improve the success rate of virtual screening in these scenarios, as well as to reduce human intervention, with the goal of streamlining early-stage drug discovery. DockBox is an earlier example of the tools that we are trying to develop. We are particularly interested in developing strategies suitable for “undruggable” targets such as protein-protein interaction sites, which are rapidly gaining attention in many therapeutic areas such as cancer and viral diseases.
Computational strategies to overcome cancer drug resistance
We use computational methods to study proteins that are responsible for drug resistance in cancer cells. We are particularly interested in investigating the structure-function relationship of DNA repair proteins. It has been established that chemical modulation of their activity in cancer cells improves the response to DNA-damaging chemotherapies such as platinum-based drugs. Thus, we employ virtual screening to discover new small molecules inhibitors of these proteins, with the goal of opening new venues for developing more effective combination cancer therapies. For example, we have previously demonstrated the effectiveness of targeting the dimerization interface of the ERCC1-XPF endonuclease with small molecules in order to sensitize cancer cells to alkylating agents.
